Time Travel



Original Source Here

Handy Kurniawan, Tamazi Sepiashvili, Shumpei Morimoto
Supervisor: Oriol Corcoll Andreu

Introduction

At first glance, the title of the project might seem some sort of fantasy and for now an impossible topic to talk about. You are mostly right, but hey, have you heard about Reinforcement Learning? Our team has gathered to explore the possibilities of Deep Learning and examine if an AI agent can learn from “mistakes” made in past.

In this project, we are creating a custom environment. The environment is trivial and provides an agent with defined rules and design of a “game”. Here we have starting and ending points. The aim of the AI is to get to the end and with the help of Reinforcement Learning, possibly go several decisions back and improve its performance. What we eventually want to imitate is how the human thinks when making decisions from the experiences based on the reward signal. As might be expected, our model will be designed with the help of a rewarding system, which means there will be a reward score for the agent with every decision it makes. This will be a measure of how much the AI is motivated to make a similar decision. All of the methods and models will be described in the following sections.

Methodology

Let’s start with the description of the Reinforcement Model concept.

Reinforcement Learning

Reinforcement learning is a machine learning technology that enables an agent to learn in an interactive environment by trial and error using feedback from its actions and experiences. It is employed by various software and machines to find the best possible behavior or path it should take in a specific situation. In our project, we will give a reward to the agent for each step so that the agent can reach the goal on a better path. You can view the general design of the reinforcement learning method in Figure 1.

Figure 1. Reinforcement Learning Model

AI/ML

Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot

%d bloggers like this: